Digital Resources for Reading Japanese Seals

Although resources for learning to read or assist with deciphering cursive Japanese script found in historical books and documents are fairly accessible—including courses, books, websites, and various digital tools, some of which we have previously covered in The Digital Orientalist—there seems to be a paucity of resources for learning how to decipher seals or stamps. In fact, many scholars and students I know, including myself, struggle with this task.

Those who have come across a seal in a Japanese text and searched for some way to decipher it online have undoubtedly come across the Seal Script Dataset (J. Tensho jitai dētasetto 篆書字体データセット). The dataset was collected by the National Institute of Japanese Literature (J. Kokubungaku kenkyū shiryōkan 国文学研究資料館) in a project led by Aota Sumi 青田寿美 and is currently hosted on the website of the ROIS-DS Center for Open Data in the Humanities (CODH). After accessing the website one can perform a search for characters using in the dataset in the same manner that one might use the Cursive Character Database Search (J. Kuzushiji dētabēsu kensaku くずし字データベース検索) system. The user inputs the character they wish to search into the search engine and will be presented with images from the dataset matching that character that they can compare to the one in their document or text. It works well, but there are some things that users should be aware of. First, whilst this may be obvious to some, seal or sigillary script (J. tenshotai 篆書体) does not refer to “characters used on seals” in general but to a specific script type. The characters that appear on stamps include a variety of styles of which seal script is just one. The system, therefore, isn’t particularly useful if your stamp is written in other writing styles. Second, and whilst this is also likely obvious, the coverage is limited by the texts used to create the dataset. This dataset consists not of stamps themselves, but of mostly dictionary-style resources for reading seal script. A third thing users should be aware of is that, while the Cursive Character Database requires searches using only modern forms (J. shinjitai 新字体)—while also including old forms (J. kyūjitai 旧字体) in the data for their modern counterparts—the same does not apply to the Seal Script Dataset, which uses old forms for searching instead. For example, the word kuni 国 produces results in the Cursive Character Database, but not in the Seal Script Dataset where one must search for the character kuni 國.

Let’s see how this works with an example. Above is the image of a seal from a publisher Kiseidō which consists of the characters 龜生堂. Versions of each character can be found by searching the dataset, however, where one might run into a problem is with the character kame 龜 especially if one is used to transcribing into modern forms of characters or using the Cursive Character Database. If we type in the modern kame 亀 instead, we get no results. Clicking on a result shows us that character in its source text.

Another system that users may be aware of is the Ownership Stamp Database (J. Zōshoin dētabēsu kensaku shisutemu  蔵書印データベース検索システム) created by the International Institute for Digital Humanities. This database includes the same type of search function, however, search parameters can be customized to a much greater degree. For example one can search for the text in the seal, by the seal’s owner, the shape of the seal, the style of the script used in the seal etc. Search results also supply this and other potentially useful information such as stamp color. Users can also search according to both new and old character forms though the results are different. For example, kame 亀 produces 108 results whereas kame 龜produces only 45. Nevertheless, as can be observed in the below screenshot some of the results in a search for kame 亀 actually show the character kame 龜. This might be a result of the way in which the data has been labelled. There are potential advantages to this—it is easy to overlook old forms when performing searches—however, an approach which combines new and old characters as adopted by CODH potentially safeguards against missing relevant results. In addition to all this and as illustrated in the below screenshot, the results consist of whole stamps rather than individual characters. This is extremely helpful since we can see the characters in situ helping with identification. Clicking on the image of stamp provides additional information about it.

The reader may have noticed that the search functions that I have discussed so far actually don’t help us to decipher characters, but simply confirm suspicions we might already have about the identity of a character in a seal. To rephrase things, I am very unlikely to discover the characters in the above seal (龜生堂) without having some idea about their identity to begin with. The Ownership Stamp Database offers some tools that aid in what we might term “deciphering” (as opposed to “confirming”) a character. Its Seal Script Radical Search System (J. Tenji bushu kensaku shisutemu 篆字部首検索システム) allows users to search for characters according to radical using data from the Seal Script Dataset. As such, if the user is able to recognize part of the character in question they may be able to identify it using this search method. For example, if we recognized that the final character in our stamp (dō 堂) contains the radical tsuchi 土, but are unable to identify the character as a whole we may be able to find it by searching for characters that contain this radical. The result is pictured in the below screenshot. From here we can quickly navigate to relevant sources of information and the character as featured within stamps found in the database.

The project also offers the Seal Script Search by Image (J. Tenji gazō kensaku 篆字画像検索) tool. This tool allows users to search for characters in a specific image and works on the user’s web browser on their computer or mobile phone. The user uploads a screenshot or takes a photo of the stamp or seal script character they are working on and uploads it. The user then crops the image to the size of one character as necessary and clicks on the analyze button. The cropping function is a bit difficult to use on mobile phone since it is fiddly, but it functions fine on a computer. The platform supplies potential characters and suggested probabilities for each, as well as links to the characters in stamps in the database, and the characters in the Seal Script Dataset. I haven’t used this part of the platform extensively enough to give a verdict, but it seems that it can be difficult to crop a single character depending on the nature of the scan or photograph which often leads to an error when trying to analyze the characters. Nevertheless, it has the potential to be extremely useful for those stuck deciphering a stamp.

A final method for helping to confirm the contents of a stamp should also be mentioned. Various stamp makers offer stamp preview services that can be used when creating a stamp. Inkan no takumi 印鑑の匠 offers one such service. Potential customers can choose different patterns and different styles of writing to create the preview of a modern-day name stamp. It was surprisingly easy to create a stamp remarkably similar to the one that appears in our example (see image below). This might form a final stage of checking if one has deciphered a stamp correctly or offer the chance to have a bit of fun with the stamps that one comes across. Of course, the results here are likely to be mixed since it these preview systems are not designed for the work that we historians and literary scholars are conducting, but rather to sell stamps to customers in 21^st century Japan.

Although there aren’t a great deal of tools to help with deciphering stamps or confirming the characters that appear in them, those tools that are available offer ample assistance. It seems that the number of tools in this area have been growing rapidly over the past few years so we might expect more to emerge in the future. I hope this brief survey of some of the tools available will help those who, like me, struggle when it comes to reading the stamps that appear in historical Japanese sources.